Software metrics' research has produced as many failures as successes. Theoretical results supported by application experiences help us understand many of the pitfalls. We will discuss how some of these pitfalls can be avoided and how metrics' research can be made more robust so that we can improve the measurements.